Merge pull request #2377 from pipecat-ai/filipi/fast_api_freeze_issue

Fixed an issue in BaseOutputTransport where the loop could consume all CPU.

CHANGELOG.md

@@ -81,6 +81,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- Fixed an issue in the `BaseOutputTransport`, mainly reproducible with
+  `FastAPIWebsocketOutputTransport` when the audio mixer was enabled, where the
+  loop could consume 100% CPU by continuously returning without delay, preventing
+  other asyncio tasks (such as cancellation or shutdown signals) from being
+  processed.
+
 - Fixed an issue where `BotStartedSpeakingFrame` and `BotStoppedSpeakingFrame`
   were not emitted when using `TavusVideoService` or `HeyGenVideoService`.
 

src/pipecat/transports/base_output.py

@@ -658,6 +658,11 @@ class BaseOutputTransport(FrameProcessor):
                             num_channels=self._params.audio_out_channels,
                         )
                         yield frame
+                        # Allow other asyncio tasks to execute by adding a small sleep
+                        # Without this sleep, in task cancellation scenarios, this loop would
+                        # continuously return without any delay, leading to 100% CPU utilization
+                        # and preventing cancel/stop signals from being processed properly
+                        await asyncio.sleep(0)
 
             if self._mixer:
                 return with_mixer(BOT_VAD_STOP_SECS)

src/pipecat/transports/network/fastapi_websocket.py

@@ -419,12 +419,7 @@ class FastAPIWebsocketOutputTransport(BaseOutputTransport):
         Args:
             frame: The output audio frame to write.
         """
-        if self._client.is_closing:
-            return
-
-        if not self._client.is_connected:
-            # Simulate audio playback with a sleep.
-            await self._write_audio_sleep()
+        if self._client.is_closing or not self._client.is_connected:
             return
 
         frame = OutputAudioRawFrame(

src/pipecat/transports/network/websocket_server.py

@@ -353,8 +353,6 @@ class WebsocketServerOutputTransport(BaseOutputTransport):
             frame: The output audio frame to write.
         """
         if not self._websocket:
-            # Simulate audio playback with a sleep.
-            await self._write_audio_sleep()
             return
 
         frame = OutputAudioRawFrame(